[llvm] [AMDGPU] Account for inline asm size in inst_pref_size calculation (PR #192306)

[Matt Arsenault via llvm-commits]

================
@@ -232,13 +232,61 @@ void AMDGPUAsmPrinter::emitFunctionBodyStart() {
     HSAMetadataStream->emitKernel(*MF, CurrentProgramInfo);
 }
 
+/// Set bits in a kernel descriptor MCExpr field:
+///   return ((Dst & ~Mask) | (Value << Shift))
+static const MCExpr *setBits(const MCExpr *Dst, const MCExpr *Value,
+                             uint32_t Mask, uint32_t Shift, MCContext &Ctx) {
+  const auto *Shft = MCConstantExpr::create(Shift, Ctx);
+  const auto *Msk = MCConstantExpr::create(Mask, Ctx);
+  Dst = MCBinaryExpr::createAnd(Dst, MCUnaryExpr::createNot(Msk, Ctx), Ctx);
+  Dst = MCBinaryExpr::createOr(Dst, MCBinaryExpr::createShl(Value, Shft, Ctx),
+                               Ctx);
+  return Dst;
+}
+
 void AMDGPUAsmPrinter::endFunction(const MachineFunction *MF) {
   const SIMachineFunctionInfo &MFI = *MF->getInfo<SIMachineFunctionInfo>();
   if (!MFI.isEntryFunction())
     return;
 
   assert(TM.getTargetTriple().getOS() == Triple::AMDHSA);
 
+  const GCNSubtarget &STM = MF->getSubtarget<GCNSubtarget>();
+  MCContext &Ctx = MF->getContext();
+
+  AMDGPU::MCKernelDescriptor KD =
+      getAmdhsaKernelDescriptor(*MF, CurrentProgramInfo);
+
+  // Compute inst_pref_size using MCExpr label subtraction for exact code
+  // size. At this point .Lfunc_end has been emitted (by the base AsmPrinter)
+  // right after the function code, so (Lfunc_end - func_sym) gives the
+  // exact function code size in bytes.
+  // We store it as a separate KD field rather than OR'ing into
+  // compute_pgm_rsrc3, because the label subtraction MCExpr is unresolvable
+  // in text mode and would prevent printing of other fields (e.g.
+  // named_barrier_count) that share the same register.
+  if (isGFX11Plus(STM)) {
+    const MCExpr *CodeSizeExpr = MCBinaryExpr::createSub(
+        MCSymbolRefExpr::create(getFunctionEnd(), OutContext),
+        MCSymbolRefExpr::create(CurrentFnSym, OutContext), OutContext);
+
+    uint32_t Width;
+    if (isGFX11(STM)) {
----------------
arsenm wrote:

Same, also should really bee a hasInstPrefSize 

https://github.com/llvm/llvm-project/pull/192306